DATAKICKS¶
INTRODUCTION¶
Reason behind selecting FIFA as our topic :¶
Our group, Datakicks will be performing data analysis on the popular video-game FIFA 20. EA’s FIFA is built on an arsenal of real-world data. Professional soccer consists of 800 teams (comprised of 18,000 players) across a variety of countries, and each is represented in-game. Our dataset includes information on players’ background such as country of origin, club team height, weight, age and position as well as their stats from the games such as number of completed passes and their preferred foot to use while playing. Given this dataset, we are able to not only analyze players information on a national level but on a club level as well within the soccer league industry.¶
Other than all group members having a liking towards some sport related topic (especially football), it was a chance for us to couple this interest with python skills and develop an ideal dream team. Every other thing then just followed and along the way, we were able to proceed with intensive scraping and cleansing and further develop meaningful and attractive visualizations too (as stated in peer feedback).¶
In simple words, it is an attempt to create a treasure of information and stats for all the FOOTBALL LOVERS ⚽️ !!!!¶
Stepwise implementation of acquiring data:¶
scrapy startproject command was used on terminal to create a project with default files such as items.py, middlewares.py, pipelines.py and settings.py along with a folder named spiders¶
This is where our .py script is present which contains spider named 'fifadatacrawler' to crawl the sofifa.com website¶
Following is the screenshot of the entire code which will make it more clear and concise¶
In [1]:
from IPython.display import Image
from PIL import Image
.py script for the spider to crawl the mentioned url and extract data¶
In [2]:
path="/Users/vats/Downloads/1stscraping.PNG"
display(Image.open(path))
In [3]:
path="/Users/vats/Downloads/2ndscraping.PNG"
display(Image.open(path))
Code explanantion - keeping it as simple as possible¶
Why used scrapy over BeautifulSoup ? This is because we had the chance to work on BeautifulSoup while in class and even during homework asssignments and hence we decided to tap an unexplored territory and develop more skills.¶
Firstly importing all modules used throughout, followed by defining class which has the spider named 'fifadatacrawler' which is the heart of our project.¶
start_requests function has list of urls from where the spider can scrape data but in our case it is just one website 'sofifa.com'. It further makes a callback to next function defined within same class named parse.¶
This is the place where we made use of our HTML knowledge of xpath notations and css locators to acquire data on each football player listed on the website.¶
You might see 62 key value pairs getting appended to dictionary in one iteration of for loop where each row of code corresponds to the exact player feature for which data is needed.¶
Last three lines of code is telling the spider to continue to the next webpage and repeat the same process¶
This is how we managed to scrape a huge dataset of 20100 rows and 62 columns which has then been reduced and formatted explained in the Data Cleaning section of this ipynb.¶
Screenshots of commands on terminal for spider to start crawling.¶
In [4]:
path="/Users/vats/Downloads/scrapyterminal1.JPG"
display(Image.open(path))
In [5]:
path="/Users/vats/Downloads/scrapyterminal2.JPG"
display(Image.open(path))
In [6]:
path="/Users/vats/Downloads/scrapyterminal4.png"
display(Image.open(path))
In [7]:
path="/Users/vats/Downloads/scrapyterminal3.png"
display(Image.open(path))
We have completed scraping of player images from the website and stored the urls in a .csv format.¶
Substantial effort went into solving changes that needed to be made in default .py files when creating a project via terminal command as mentioned before.¶
We have done that separately as we didn't want to mix it up with the .csv file on which cleansing , analysis, visualization and to some extent prediction too had to be done.¶
Below is a screenshot showcasing extra efforts we have put in.¶
In [8]:
path="/Users/vats/Downloads/imagescrapingJSON.png"
display(Image.open(path))
In [9]:
#importing packages
import re
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
plt.style.use('fivethirtyeight')
df = pd.read_csv('fulldata.csv')
%pylab inline
import datetime as dt
import statsmodels.api as sm
pd.set_option('display.max_columns', 40)
Data Cleaning¶
The huge dataset with 20100 rows and 62 columns that we have extracted needs some cleaning before we go ahead with the analysis.¶
Height and weight are important characteristics of a player. These characteristics can help us draw useful insights. More about this in the analysis part.¶
Our dataset contains the numeric values of height alongwith the feet and inches symbols (eg. 6'2"). Similarly the weight column contains the numeric value along with 'lbs' (eg. 185lbs).¶
For our analysis, we only need the numeric values of these so here we define a function to extract only the numeric values of height and weight.¶
In [10]:
def getNumbers(str):
array = re.findall(r'[0-9]+', str)
return array
In [11]:
feet=[]
inches= []
for i in range(len(df)) :
list1 = getNumbers(df['Height'][i])
feet.append(list1[0])
inches.append(list1[1])
list1 = []
In [12]:
feet_df = pd.DataFrame(feet)
inches_df = pd.DataFrame(inches)
In [13]:
df['feet'] = feet_df.astype(float)
In [14]:
df['inches'] = inches_df.astype(float)
In [15]:
height_in_cms = []
for x in range (len(df)) :
h = ((df['feet'][x])*30.48 + (df['inches'][x]*2.54))
height_in_cms.append(h)
In [16]:
df['Height in cms']=height_in_cms
In [17]:
weight = []
for i in range(len(df)) :
list2 = getNumbers(df['Weight'][i])
weight.append(list2[0])
list2 = []
In [18]:
df['Weight']=weight
In [19]:
df['Value']=df['Value'].apply(lambda x: x.strip('€'))
df['Wage']=df['Wage'].apply(lambda x: x.strip('€'))
The dataset columns of Value and Wage are not consistent. Some of the values are in Millions (M) while some are in Thousands (K). Here, we have made it consistent for accuracy in our further analysis.¶
In [20]:
df.Value = (df.Value.replace(r'[KM]+$', '', regex=True).astype(float) * \
df.Value.str.extract(r'[\d\.]+([KM]+)', expand=False).fillna(1).replace(['K','M'], [10**3, 10**6]).astype(int))
df.Wage = (df.Wage.replace(r'[KM]+$', '', regex=True).astype(float) * \
df.Wage.str.extract(r'[\d\.]+([KM]+)', expand=False).fillna(1).replace(['K','M'], [10**3, 10**6]).astype(int))
Our extracted dataset is huge with 20100 rows and 62 columns. All of these columns aren't relevant and are not required for the analysis. So we go ahead with deleting the extra columns that are not required for analysis.¶
In [21]:
del df['LoanDateEnd']
del df['BestPosition']
del df['ReleaseClause']
del df['IntlReputation']
del df['Growth']
del df['BaseStats']
del df['AttackingWorkRate']
del df['DefensiveWorkRate']
del df['Hits']
Free agents are not needed for this analysis as we only want players who play for a team to be included in our analysis so we drop values that had null for club.¶
In [22]:
df=df[df['Club'].notnull()]
df=df[df['Club']!='111648']
In [23]:
#To check the null values present in columns
df.isnull().sum()
Out[23]:
In [24]:
#We check again for null values, Joined can be null as it is not important to be filled up for our analysis
mean=['Volleys','curve','Agility','Balance','Jumping','Interceptions','Positioning','Vision','Composure','SlidingTackle']
for i in mean:
df[i]=df[i].fillna((df[i].mean()))
df.isnull().sum()
Out[24]:
Data Analysis¶
Our goal in this step is of discovering useful information, informing conclusion and supporting decision-making related to our data set. Data analysis has multiple facets and approaches, encompassing diverse techniques under a variety of names, and is used in different business, science, and social science domains. In today's business world, data analysis plays a role in making decisions more scientific and helping businesses operate more effectively¶
The analysis is divided into 7 important questions that we wanted to get out of the data set¶
1) Do physical attributes affect the overall of the player?¶
2) Who are the most loyal players in the game and what is the biggest age gap?¶
3) Which is the least popular position and why?¶
4) Is there a relation between the wage and value of players?¶
5) Is there a difference between the highest rated players and the dream team?¶
6) How is the value of the player affected by his potential?¶
7) Which are the most lethal countries in the game?¶
Analysis 1 - Physical Attributes¶
Do physical attributes affect the overall of the player and is there any kind of physical feature that affects the overall of a player?¶
In [25]:
#Is there a relation between age and overall rating?
plt.figure(1,figsize=(15,5))
sns.regplot(df['Age'],df['Overall'])
plt.title('Age vs Overall Scatter Plot')
plt.show()
#Is there a relation between height and overall rating?
plt.figure(1,figsize=(10,5))
sns.regplot(df['Height in cms'],df['Overall'])
plt.title('Height in cms vs Overall Scatter Plot')
plt.show()
Inference¶
From the trend line in the scatterplots displayed above, it is observed that, age significantly affects the overall rating of the player positively as there is a clear upward trend. This validates the fact that experience increases with age since players acquire more skills through training and playing for more number of years. On the other hand, height does not really seem to affect the overall rating of the player as this trend line seems to be flat. Thus height is not an important attribute to consider while valuing a player.¶
Analysis 2 - Age and Loyalty¶
Who are the most loyal players in the games and what is the biggest age gap in the game?¶
In [26]:
#The most loyal players
#Finding the number of years the players have played for the same team and then ordering them by the highest value
now = datetime.datetime.now()
df['Year of Joining'] = df.Joined.dropna().map(lambda x: x.split(',')[1].split(' ')[1])
df['Loyalty'] = (df['Year of Joining'].dropna().map(lambda x: now.year - int(x))).astype('int')
Loyalty = df[['Name', 'Club', 'Age','Joined','Loyalty']].sort_values(by = 'Loyalty', ascending = False).head(10)
Loyalty.set_index('Name', inplace=True)
Loyalty
Out[26]:
The most loyal players¶
Clubs with the highest age gap between players¶
In [27]:
#Clubs with the highest age gap between players
def agegap(group):
return group.max() - group.min()
df.groupby(by=df['Club'])['Age'].agg(agegap).sort_values(ascending=False).head()
Out[27]:
Inference¶
The analysis shows us retention on how passionate a player can be about their club and football by remaining with the same football club for their career. In addition, from the visualization one can see that a player can be younger but have more loyalty points than someone who is older than than them. It also tells us that it's never too early to start playing football and never too early to stop either.¶
All it takes is passion, will and determination from the player themselves :)¶
Analysis 3 - Positions¶
Which is the least popular position and why?¶
In [28]:
#Finding the distribution of players in each position
plt.figure(figsize = (12, 8))
sns.set(style = 'dark', palette = 'deep', color_codes = True)
ax = sns.countplot('Position', data = df, color = 'violet')
ax.set_xlabel(xlabel = 'Different Positions in Soccer', fontsize = 14)
ax.set_ylabel(ylabel = 'Count of Players in each Position', fontsize = 14)
ax.set_title(label = 'Graph of Positions and Players', fontsize = 20)
plt.show()
Conclusion/Inference¶
This analysis shows us that the centerback position is the most popular position to play in football, whereas it is very rare to find a player playing Right Forward 'RF', this might be due to the fact that it may be difficult to adapt to the skills that the position requires or that there are not enough teams playing with a formation that uses right forward in competition.¶
Analysis 4 - Wage vs Value of a player¶
Is there a relation between the wage and value of players?¶
First, lets look at some statistics of the wages of players in the game¶
Teams paying the most in wages¶
In [29]:
#Teams paying the most wages
df.groupby(by=df['Club'])['Wage'].agg(sum).sort_values(ascending=False).head()
Out[29]:
Players who earn the most wages and the least wages in football¶
In [30]:
#Players who earn the most
df.sort_values(by = 'Wage' , ascending = False)[['Name','Club','Country','Overall','Age','Position','Value','Wage']].head()
Out[30]:
Lets also look at some statistics of the value of players in the game¶
The most expensive players¶
In [31]:
#The most expensive players
df.sort_values(by = 'Value' , ascending = False)[['Name','Club','Country','Overall','Age','Position','Value','Wage']].head()
Out[31]:
In [32]:
np.corrcoef(df['Value'],df['Wage'])
Out[32]:
In [33]:
plt.figure(1,figsize=(10,5))
sns.regplot(df['Wage'],df['Value'])
plt.title('Value vs Wage Scatter Plot')
plt.show()
Analysis 5 - Ratings, Highest rated Team vs Dream Team¶
Is there a difference between the highest rated players and the dream team?¶
Let's look at the highest rated players and team and whether they correspond to the dream team¶
In [34]:
#Highest rated player in each position
df.groupby(by=df['Position'])['Overall'].agg('max')
x=pd.DataFrame()
grouped = df.groupby(by=df['Position'])
position=df['Position'].unique()
for i in position:
x=x.append(grouped.get_group(i).sort_values(by='Overall',ascending=False).head(1),ignore_index=True)
x
Out[34]:
These are the players who are the highest rated in each position.¶
Let's now create the Dream Team by finding 10 attributes or features which are most important for each position and see if there is a difference¶
In [35]:
# What are the best 10 features for each position and according to that we make our dream team
dream_team={}
features = ('Crossing', 'finishing',
'HeadingAccuracy', 'ShortPass', 'Volleys', 'Dribble',
'curve', 'FreeKickAcc', 'LongPass', 'BallControl',
'Acceleration', 'SprintSpeed', 'Agility', 'Reactions', 'Balance',
'ShotPower', 'Jumping', 'Stamina', 'Strength',
'LongShots', 'Aggression', 'Interceptions',
'Positioning', 'Vision', 'Penalties', 'Composure',
'Marking', 'StandingTackle', 'SlidingTackle', 'GKdiving',
'GKhandling', 'GKkicking', 'GKpositioning', 'GKreflexes')
for i, val in df.groupby(df['Position'])[features].mean().iterrows():
dream_team[i]=list(val.nlargest(10).index)
print('For {}: {}, {}, {}, {}, {},{},{},{},{},{}'.format(i, *list(val.nlargest(10).index)))
In [36]:
df2=df.copy()
def sum_frame_by_column(frame, new_col_name, list_of_cols_to_sum):
frame[new_col_name] = frame[list_of_cols_to_sum].astype(float).sum(axis=1)
return(frame.sort_values(by=new_col_name,ascending=False).head(1))
best11=pd.DataFrame()
In [37]:
#For GK
df_GK=df[df['Position']=='GK']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_GK, 'Sum', ['GKreflexes', 'GKdiving', 'GKhandling', 'GKpositioning', 'GKkicking','Strength','Jumping','Reactions','Balance','Composure'])
best11=best11.append(sum_frame_by_column(df_GK, 'Sum', ['GKreflexes', 'GKdiving', 'GKhandling', 'GKpositioning', 'GKkicking','Strength','Jumping','Reactions','Balance','Composure']),ignore_index=True)
best11
Out[37]:
In [38]:
#For LB
df_LB=df[df['Position']=='LB']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_LB, 'Sum', ['Strength', 'Jumping', 'StandingTackle', 'Aggression', 'SlidingTackle','HeadingAccuracy','Marking','Interceptions','Stamina','Reactions'])
best11=best11.append(sum_frame_by_column(df_LB, 'Sum', ['Strength', 'Jumping', 'StandingTackle', 'Aggression', 'SlidingTackle','HeadingAccuracy','Marking','Interceptions','Stamina','Reactions']),ignore_index=True)
best11
Out[38]:
In [39]:
#For the 2 CBs
df_CB=df[df['Position']=='CB']
pd.set_option('mode.chained_assignment', None)
def sum_frame_by_column2(frame, new_col_name, list_of_cols_to_sum):
frame[new_col_name] = frame[list_of_cols_to_sum].astype(float).sum(axis=1)
return(frame.sort_values(by=new_col_name,ascending=False).head(2))
sum_frame_by_column2(df_CB, 'Sum', ['Acceleration', 'SprintSpeed', 'Stamina', 'Balance', 'Agility','Jumping','StandingTackle','SlidingTackle','Aggression','Crossing'])
best11=best11.append(sum_frame_by_column2(df_CB, 'Sum', ['Acceleration', 'SprintSpeed', 'Stamina', 'Balance', 'Agility','Jumping','StandingTackle','SlidingTackle','Aggression','Crossing']),ignore_index=True)
best11
Out[39]:
In [40]:
#For RB
df_RB=df[df['Position']=='RB']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_RB, 'Sum', ['Acceleration', 'SprintSpeed', 'Stamina', 'Balance', 'Agility','Jumping','StandingTackle','SlidingTackle','Aggression','Strength'])
best11=best11.append(sum_frame_by_column(df_RB, 'Sum', ['Acceleration', 'SprintSpeed', 'Stamina', 'Balance', 'Agility','Jumping','StandingTackle','SlidingTackle','Aggression','Strength']),ignore_index=True)
best11
Out[40]:
In [41]:
#For the 2 CMs
df_CM=df[df['Position']=='CM']
pd.set_option('mode.chained_assignment', None)
def sum_frame_by_column2(frame, new_col_name, list_of_cols_to_sum):
frame[new_col_name] = frame[list_of_cols_to_sum].astype(float).sum(axis=1)
return(frame.sort_values(by=new_col_name,ascending=False).head(2))
sum_frame_by_column2(df_CM, 'Sum', [ 'Balance', 'ShortPass', 'Agility', 'Stamina', 'Acceleration','BallControl','LongPass','SprintSpeed','Dribble','Vision'])
best11=best11.append(sum_frame_by_column2(df_CM, 'Sum', ['Balance', 'ShortPass', 'Agility', 'Stamina', 'Acceleration','BallControl','LongPass','SprintSpeed','Dribble','Vision']),ignore_index=True)
best11
Out[41]:
In [42]:
#For LW
df_LW=df[df['Position']=='LW']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_LW, 'Sum', ['Acceleration', 'SprintSpeed', 'Agility', 'Balance', 'Dribble','BallControl','ShotPower','ShortPass','Positioning','Stamina'])
best11=best11.append(sum_frame_by_column(df_LW, 'Sum', ['Acceleration', 'SprintSpeed', 'Agility', 'Balance', 'Dribble','BallControl','ShotPower','ShortPass','Positioning','Stamina']),ignore_index=True)
best11
Out[42]:
In [43]:
#For CAM
df_CAM=df[df['Position']=='CAM']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_CAM, 'Sum', ['Balance', 'Agility', 'Acceleration', 'SprintSpeed', 'BallControl','Dribble','ShortPass','Vision','ShotPower','Composure'])
best11=best11.append(sum_frame_by_column(df_CAM, 'Sum', ['Balance', 'Agility', 'Acceleration', 'SprintSpeed', 'BallControl','Dribble','ShortPass','Vision','ShotPower','Composure']),ignore_index=True)
best11
Out[43]:
In [44]:
#For RW
df_RW=df[df['Position']=='RW']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_RW, 'Sum', ['Acceleration', 'SprintSpeed', 'Agility', 'Balance', 'Dribble','BallControl','ShotPower','Stamina','Positioning','Crossing'])
best11=best11.append(sum_frame_by_column(df_RW, 'Sum', ['Acceleration', 'SprintSpeed', 'Agility', 'Balance', 'Dribble','BallControl','ShotPower','Stamina','Positioning','Crossing']),ignore_index=True)
best11
Out[44]:
In [45]:
#For ST
df_ST=df[df['Position']=='ST']
pd.set_option('mode.chained_assignment', None)
sum_frame_by_column(df_ST, 'Sum', ['SprintSpeed', 'Acceleration', 'Strength', 'Jumping', 'finishing','ShotPower','Agility','Positioning','Balance','BallControl'])
best11=best11.append(sum_frame_by_column(df_ST, 'Sum', ['SprintSpeed', 'Acceleration', 'Strength', 'Jumping', 'finishing','ShotPower','Agility','Positioning','Balance','BallControl']),ignore_index=True)
best11
Out[45]:
Our dream team is given by¶
In [46]:
best11
Out[46]:
In [47]:
#This is the dream team line up
path="/Users/vats/Downloads/dreamteam.PNG"
display(Image.open(path))
This is the team which the best team based on the quality or the attributes which are required for each position¶
Inference¶
The analysis shows us that Dream Team is not equivalent to the highest rated team, this is because for each position there are certain attributes that are important. For instance, for a centerback, the position requires a player who has good positioning and defending skills to prevent scoring as supposed to a center forward whose focus is more on shooting accuracy. Therefore, the Dream Team is made according to the most important features for each position and not the rating per player.¶
Analysis 6 - Value and Potential¶
How is the value of the player affected by his potential?¶
Let's find the clubs with the highest potential and most valuable teams and see if there are any correlations to it¶
In [48]:
#Most valuable teams
df.groupby(by=df['Club'])['Value'].agg(sum).sort_values(ascending=False).head()
Out[48]:
In [49]:
#Clubs with the highest mean potential
df.groupby(by=df['Club'])['Potential'].agg('mean').sort_values(ascending=False).head()
Out[49]:
The above analysis shows that the highest potential teams are also the most valuable teams, so there has to be some kind of relation between the two terms.¶
Potential decides what overall a player will be able to reach after a few seasons, although players can, of course, exceed or fall short of this target. You can get an idea of a player's potential through little remarks on their scout report.¶
"Showing great potential" means a player has potential of 80-84; "an exciting prospect" means a player has potential of 85-89; and "has the potential to be special" means a player has potential of 90 plus.¶
We decided to select potential as the independent variable to check how much does the future rating of a player(ie potential) affect the value of a player if any, let's further examine that using a regression model¶
Let's create a regression model created to analyze how the value of a player is affected by the potential of a player¶
In [50]:
col=['Potential']
X = df[col]
X = sm.add_constant(X)
y = df['Value']
In [51]:
pd.concat([X, y], axis=1).head(10)
Out[51]:
In [52]:
model = sm.OLS(endog=y, exog=X)
results = model.fit()
results.summary()
Out[52]:
Inference¶
This analysis shows us that on average an increase in 1 unit of potential for a player,his value increases by 479,500 when all other factors are constant. Thus, we can conclude there is a high positive correlation between potential and value.¶
The more expensive option for a winger is the aforementioned Jadon Sancho. The Borussia Dortmund man starts off at a much higher rating of 78 and also has an exceptional potential of 90 overall. On top of that, his in-game stats are excellent with five-star skills complemented by 88 acceleration, 87 sprint speed, 90 agility and 87 dribbling. Despite all this, Sancho is still relatively affordable with a starting transfer value of around £17m. If you have the extra budget, he is well worth picking up.¶
Which are the most lethal countries in the game?¶
Here the focus shifts to showcasing visualization skill set and generate a world map with each country's overall rating (based on player features) highlighted.¶
In [53]:
# different set of packages have to be installed and imported for this map
import plotly.offline as py
from plotly.offline import init_notebook_mode, iplot
init_notebook_mode(connected = True) #initiate Plotly notebook mode
import plotly.graph_objs as go
In [54]:
overall = pd.DataFrame(df.groupby(['Country'])['Overall'].mean().reset_index())
count = pd.DataFrame(overall.groupby('Country')['Overall'].mean().reset_index())
#this count dataframe has countrywise mean overall value (considered overall of every player for a country and then computed mean)
In [55]:
trace = [go.Choropleth(
colorscale = 'YlGnBu',
locationmode = 'country names',
locations = count['Country'],
text = count['Country'],
z = count['Overall'],
)]
layout = go.Layout(title = 'Country vs Mean Overall Rating')
fig = go.Figure(data = trace, layout = layout)
py.iplot(fig)
Inference¶
This map is interactive in nature and when we hover over any particular region, it displays the mean overall value(all player features included) along with the country's name. Due to this clean visualization, it is evident that regions with dark blue color have a higher overall value and as the shade moves towards yellow , there is a decrease observed in overall value.¶
Bottom line is that we can point out national teams with good quality players and even compare a country's standing with respect to others.¶
Conclusion of the above analysis¶
1) We wanted to analyze the physical attributes and how they affect the rating of a player¶
2) We analyzed the age gap between players and loyalty of players¶
3) We found the distribution of players in each position and made inferences¶
4) There was a relation between the wage and value of players¶
5) We found a difference between the highest rated players and the dream team¶
6) We observed the value of the player being affected by his potential¶
7) We mapped the countries of players, with their mean rating to find out the most lethal countries¶
Final Conclusion¶
For this project, we were able to gather several analyses and generate visualizations from sofifa.com. FIFA statistics help provide analyses such as players' stats and demographics, team rankings, and an overview of the football industry as a whole. The data used can not only just capture individual as well as team statistics, yet when reviewed later the data helps highlight strengths, weaknesses, scout talent, and keep records of progess made over time. Statistics can also help coaches measure the effectiveness of their practice sessions and plans for future sessions to improve to development of team.¶
Overall, in terms of our dataset, the FIFA statistics used will allowed us to gauge a better understanding of player information. From our analyses we were able to understand correlations between player demographics, assess player loyalty and team retention rates, assess players' posiitoning and skills needed to complie the ideal Dream Team , and lastly were able to create visualizations to map the ratings of players across the globe. Compiling these analyses overall led to a better understanding of the football industry to analyze the player's skills and positioning individually and across teams as well. After completing this project, we have developed essential python skills with scraping and cleaning data to generating a variety of visualizations. Not only did this help us improve essential skills for data science priniciples but also allowed us to create analyses based on real world data to the football industry in which we all have high interest. Although this project , required a lot of time and effort, our team DataKicks has effectively complied analyses on the data we have collected for player information in FIFA.¶
APPENDIX¶
Choropleth map (Analysis Point 7) but using geopandas (not interactive)¶
In [56]:
import geopandas
fp = '/Users/vats/Downloads/World_Countries/World_Countries.shp'
geo_df = geopandas.read_file(fp) # this dataframe is GEOdataframe and not a normal dataframe
geo_df.plot()
Out[56]:
In [57]:
# join our dataframe named count defined in analysis point 7 with geodataframe obtained above
merged = geo_df.set_index('COUNTRY').join(count.set_index('Country'))
In [58]:
variable = "Overall" # column to visualize on the choropleth map generated
vmin, vmax = 10000,200000
fig, ax = plt.subplots(1, figsize=(10, 6))
ax.axis("off")
ax.set_title('Overall rating for each country', fontdict={'fontsize': '25', 'fontweight' : '3'})
sm = plt.cm.ScalarMappable(cmap='YlGnBu', norm=plt.Normalize(vmin=vmin, vmax=vmax)) #colorbar that you see on the right as a legend to the map
sm._A = []
cbar = fig.colorbar(sm)
merged.plot(column=variable, cmap="YlGnBu", linewidth=0.8, ax=ax, edgecolor="0.8")
Out[58]:
